AfriScience-MT (Zero-Shot): leaderboard
Translation of plain-language scientific summaries between English and six African languages (Amharic, Hausa, Luganda, Northern Sotho, Yoruba, isiZulu), from the professionally translated AfriScience-MT corpus (Abdulmumin et al., 2026). Sentence-level SSA-COMET on a 0-100 scale averaged over the 12 directions of the test split, with no in-context examples and no fine-tuning; higher is better.
Metric: SSA-COMET (0-100; sentence-level, mean over the 12 English<->African directions of the test split; zero-shot: vanilla 0-shot prompt for LLMs, direct decoding for NLLB and M2M100). Source: arxiv.org. Saturation forecast: Around February 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 65 | #121 |
| 2 | Gemini 1.5 Flash | 58.7 | #1671 |
| 3 | GPT-4o | 58 | #741 |
| 4 | Gemini 3.1 Flash Lite | 57 | #363 |
| 5 | Gemma 2 9B (IT) | 39.4 | #2275 |
| 6 | Tiny Aya Global | 36.3 | #6489 |
| 7 | Llama 3.1 8B Instruct | 34.9 | #4277 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=afriscience-mt-zero-shot · How It Works · Data refreshed daily, snapshot 2026-11-12.